A fan bearing fault diagnosis method based on incremental learning and rule reasoning
Through incremental learning and rule reasoning, combined with vibration sensors and wireless intelligent gateways, the problem of insufficient data accumulation in fan bearing fault diagnosis is solved, and high accuracy and adaptability fault diagnosis is achieved, reducing the cost of production line downtime.
Patent Information
- Application Number
- CN202310792515.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-30
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2043-06-30
AI Technical Summary
Traditional fan bearing fault diagnosis algorithms based on expert systems or empirical models cannot effectively solve the problems of insufficient data accumulation and uneven model training, resulting in a low detection rate of early and mid-term faults, which leads to huge cost of production line shutdown.
Using the method based on incremental learning and rule reasoning, signals are collected through vibration sensors, data preprocessing and feature extraction are performed using wireless intelligent gateways and controllers, and the incremental learning model and multi-result fusion decision unit are combined to achieve continuous update of the model and Converged decisions on diagnostic results.
It improves the accuracy, accuracy and comprehensiveness of fault diagnosis, enhances the adaptability and reliability of diagnosis, and avoids the shortcomings and misjudgment of a single diagnosis result.
Smart Images
Figure CN116839909B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a fan bearing fault diagnosis method based on incremental learning and rule reasoning. Background Art
[0002] At present, the field of industrial intelligent manufacturing is developing rapidly, but the technology for maintenance, analysis and diagnosis of related equipment is still immature. Some rotating equipment installation sites are often in harsh environments and still rely on maintenance personnel to use their five senses for inspection. Because the degree of faults in different periods of the fault cycle is different, the characteristic signal scale of early and mid-term faults is quite different from that of late fault signals and is easily submerged. Therefore, the detection rate of early and mid-term faults is low, and the accident cost of the entire production line being shut down due to rotating equipment failures is huge. Traditional diagnostic algorithms based on expert systems or empirical models cannot solve the problem of equipment data accumulation in the field of fault diagnosis.
[0003] A Chinese patent with publication number CN109255333B discloses a hybrid diagnosis method for rolling bearing faults of large wind turbines. First, the diagnosis results are obtained by two different types of single fault diagnosis methods, namely, a wind turbine bearing fault diagnosis method based on GVMD-SVM and a wind turbine bearing fault diagnosis method based on morphological-correlation analysis. Then, the improved DS evidence theory is used to fuse the diagnosis results of the above two methods to obtain a final more reliable diagnosis result. However, the invention is not ideal for bearing fault diagnosis with a small data base. Summary of the invention
[0004] The present invention provides a fan bearing fault diagnosis method based on incremental learning and rule reasoning. The incremental learning model is used for learning and training, which realizes continuous updating and optimization of the model, improves the adaptability and reliability of fault diagnosis, and solves the problems of very little industrial field fault data and unbalanced model training by using rule exhaustiveness, rule reasoning and incremental matching. The diagnosis results are fused and decided by a multi-result fusion decision unit, which avoids the insufficiency and misjudgment of a single diagnosis result and improves the accuracy, correctness and comprehensiveness of fault diagnosis.
[0005] In order to achieve the above object, the present invention adopts the following technical solutions:
[0006] A method for diagnosing fan bearing faults based on incremental learning and rule reasoning includes a vibration sensor, a wireless intelligent gateway and a controller. The controller includes a data acquisition unit, a vibration signal preprocessing unit, a fan fault spectrum rule reasoning unit, an incremental learning unit for fan signals and a multi-result fusion decision unit. The vibration sensor is wirelessly connected to the wireless intelligent gateway, the wireless intelligent gateway is connected to the controller, the controller collects and processes vibration signals, the fan fault spectrum rule reasoning unit performs feature quantification and rule reasoning on the data processed by the vibration signal preprocessing unit, the incremental learning unit for fan signals performs incremental learning training on the data after feature quantification and rule reasoning, and the multi-result fusion decision unit issues a diagnosis conclusion. The specific steps of the method are as follows:
[0007] (S1) collecting vibration signals of the fan and the fan motor through a vibration sensor, and uploading the vibration signals to the controller through a wireless intelligent gateway;
[0008] (S2) decoding and storing the data and cleaning the data in the controller through the data acquisition unit, and inputting the cleaned data into the vibration signal preprocessing unit for preprocessing operations such as filtering, noise reduction and eigenvalue calculation;
[0009] (S3) extracting features from the preprocessed data in a fan fault spectrum rule reasoning unit based on prior knowledge, quantifying the extracted features and enumerating rules to form rule reasoning diagnosis results and suggestions;
[0010] (S4) using the extracted features to establish an incremental learning model through an incremental learning unit of the fan signal for learning and training. When a new data type appears, new rules are automatically generated and matched and verified with the prior knowledge of bearing faults to form new diagnostic results and suggestions.
[0011] (S5) The new and old diagnosis results and suggestions of steps (S3) and (S4) are fused and decided by a multi-result fusion decision unit to output a final diagnosis conclusion.
[0012] Furthermore, in the step (S3), the extracted features are quantified, and the quantified variables include fan rotation frequency, rotation frequency energy, rotation frequency multiplication, frequency multiplication energy, bearing characteristic frequency, resonance area, resonance area low-frequency information and energy concentration area spectrum structure.
[0013] Furthermore, the incremental learning model in the step (S4) is trained using the SVDD method, and the newly added data types are identified by the kernel density estimation method.
[0014] Furthermore, the newly added learning model in the step (S4) continuously updates the new model during the learning training, and classifies and incrementally matches the quantified features of the wind turbine in a category manner.
[0015] Compared with the prior art, the present invention has the following beneficial effects:
[0016] 1) A wireless intelligent gateway and vibration sensor are used to collect the vibration signals of the fan and fan motor. Through preprocessing operations such as data cleaning, filtering and noise reduction, and eigenvalue calculation, the accuracy and reliability of the data are effectively improved, and the accuracy of fault diagnosis is improved.
[0017] 2) Based on prior knowledge, feature extraction is performed in the fan fault spectrum rule inference unit, and through rule exhaustive and quantitative processing, the problems of extremely small industrial field fault data and unbalanced model training are solved, thereby improving the accuracy of fault diagnosis.
[0018] 3) An incremental learning model is used for learning and training, the SVDD method is used to identify and classify the model, and the kernel density estimation method is used to identify new data types, thereby achieving continuous updating and optimization of the model and improving the adaptability and reliability of fault diagnosis.
[0019] 4) The diagnosis results are fused and decided by the multi-result fusion decision unit, avoiding the shortcomings and misjudgments of a single diagnosis result and improving the comprehensiveness and accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION
[0021] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings:
[0022] See Figure 1 , is a schematic diagram of the structure of the present invention. A method for diagnosing a fan bearing fault based on incremental learning and rule reasoning includes a vibration sensor, a wireless intelligent gateway and a controller, wherein the controller includes a data acquisition unit, a vibration signal preprocessing unit, a fan fault spectrum rule reasoning unit, an incremental learning unit for fan signals and a multi-result fusion decision unit, wherein the vibration sensor is wirelessly connected to the wireless intelligent gateway, the wireless intelligent gateway is connected to the controller, the controller collects and processes vibration signals, the fan fault spectrum rule reasoning unit performs feature quantification and rule reasoning on the data processed by the vibration signal preprocessing unit, the incremental learning unit for fan signals performs incremental learning training on the data after feature quantification and rule reasoning, and the multi-result fusion decision unit issues a diagnosis conclusion;
[0023] A fan bearing fault diagnosis method based on incremental learning and rule reasoning uses the vibration data of the fan and motor of the industrial heating furnace. A 3-axis acceleration vibration sensor is installed at the drive end and load end of the heating furnace fan. The specific steps are as follows:
[0024] 1) The vibration data of the heating furnace fan and motor are collected through a 3-axis acceleration vibration sensor, and the collected vibration data is transmitted to the data acquisition unit in the controller through a wireless intelligent gateway;
[0025] 2) Decoding, cleaning and saving the data in the data acquisition unit, and then performing preprocessing operations of filtering, noise reduction and eigenvalue calculation on the data through the vibration signal preprocessing unit in the controller;
[0026] 3) Based on prior knowledge, feature extraction is performed in the fan fault spectrum rule reasoning unit, and the extracted features are quantified and exhaustively enumerated. The fan rotation frequency, rotation frequency energy, rotation frequency multiplication, frequency multiplication energy, bearing characteristic frequency, resonance area, resonance area low-frequency information and energy concentration area spectrum structure are all quantified variables. Then rule reasoning is performed to exhaustively enumerate common faults of heating fans, and frequency domain rules are matched for each type of fault. Each type of fault can match multiple frequency domain rules to form diagnostic results and suggestions for rule reasoning;
[0027] 4) An incremental learning model is established through the incremental learning unit of the fan signal for learning and training. The extracted features are divided into original data types and new data types. When new data types appear, new rules are automatically generated and matched with the prior knowledge of bearing faults to form new diagnostic results and suggestions.
[0028] 5) The new and old diagnostic results and suggestions are integrated and decided through the multi-result fusion decision unit to output the final diagnostic conclusion.
[0029] [Example 1]
[0030] 1) A deep groove ball bearing 6205-RSJEMSKF was selected as the experimental object. The basic parameters of the bearing include an inner diameter of 25 mm, an outer diameter of 52 mm, and a thickness of 15 mm.
[0031] 2) Select an acceleration sensor as a vibration sensor and install it on the base shell. Upload the normal bearing data, bearing inner ring fault data, bearing outer ring fault data, bearing rolling element fault data and bearing cage fault data to the controller through the wireless intelligent gateway.
[0032] 3) Perform data preprocessing in the controller, segment the signal and extract features, and establish normal signal rules, inner ring fault rules, outer ring fault rules, rolling element fault rules and cage fault rules.
[0033] 4) Use normal data as model training data and continuously update the model through incremental learning. The feature quantity of incremental learning data gradually evolves into new categories, and the new categories are matched through rule reasoning.
[0034] 5) Gradually increase the bearing inner race fault data to a data ratio of 1:1 and match the bearing inner race fault rules.
[0035] 6) Repeat step 5, gradually adding other types of fault data to the data sample until all the data are added.
[0036] 7) With the continuous iteration and training of the model, the corresponding rules and results can be matched eventually.
[0037] 8) Experimental results show that the final accuracy of the model of this method is 98.57%
[0038] [Example 2]
[0039] The vibration data of heating furnace fans and motors used in industrial fields were selected for verification and analysis. Two heating furnace fans with exactly the same parameters were selected for testing. One fan was normal data and the other fan had an angular misalignment fault. The normal heating furnace fan was used to generate normal data training samples, and the angular misalignment heating furnace fan was used to generate fault test samples. The sampling data frequency was set to 51.2kHz, the acquisition time was set to 10s, and the signal frequency resolution was 1Hz. The abnormal signals were removed by data preprocessing methods, and the remaining data was calculated and Fourier transformed in the time domain to extract the frequency domain features. Incremental learning technology was used to continuously add training data to the model and form corresponding rules to improve the accuracy of the model. The final confirmed incremental learning model input is 1-5 times the rotation frequency and its energy relationship, the resonance zone range, the bearing characteristic frequency relationship and its harmonics, each time domain characteristic value, the low-frequency characteristics after resonance demodulation, and the signal spectrum and signal time domain waveform segmentation data.
Claims
1. A fan bearing fault diagnosis method based on incremental learning and rule reasoning, comprising a vibration sensor, a wireless intelligent gateway and a controller, characterized in that: The controller includes a data acquisition unit, a vibration signal preprocessing unit, a fan fault spectrum rule inference unit, an incremental learning unit for fan signals, and a multi-result fusion decision unit. The vibration sensor is wirelessly connected to a wireless intelligent gateway, and the wireless intelligent gateway is connected to a controller. The controller collects and processes vibration signals. The fan fault spectrum rule inference unit performs feature quantification and rule inference on the data processed by the vibration signal preprocessing unit. The incremental learning unit for fan signals performs incremental learning training on the data after feature quantification and rule inference. The multi-result fusion decision unit issues a diagnosis conclusion. The specific steps of the method are as follows: (S1) collecting vibration signals of the fan and the fan motor through a vibration sensor, and uploading the vibration signals to the controller through a wireless intelligent gateway; (S2) decoding and storing the data and cleaning the data in the controller through the data acquisition unit, and inputting the cleaned data into the vibration signal preprocessing unit for preprocessing operations such as filtering, noise reduction and eigenvalue calculation; (S3) extracting features based on prior knowledge through a fan fault spectrum rule inference unit, quantifying the extracted features and performing rule inference to generate fault diagnosis results and suggestions; (S4) Establishing an incremental learning model through the incremental learning unit of the fan signal to train the extracted features. When a new data type appears, automatically generating new rules, matching and verifying with the prior knowledge of the bearing fault, and forming new diagnostic results and suggestions; (S5) fusing the new and old diagnosis results and suggestions of steps (S3) and (S4) through a multi-result fusion decision unit and outputting a diagnosis conclusion; The incremental learning model in the step (S4) adopts the support vector data description (SVDD) method for learning and training, and identifies the newly added data type through the kernel density estimation method.
2. A fan bearing fault diagnosis method based on incremental learning and rule reasoning according to claim 1, characterized in that: In the step (S3), the extracted features are quantified, and the fan rotation frequency, rotation frequency energy, rotation frequency multiplication, frequency multiplication energy, bearing characteristic frequency, resonance area, resonance area low frequency information and energy concentration area spectrum structure are all quantified variables.
3. A method for diagnosing fan bearing faults based on incremental learning and rule reasoning according to claim 1, characterized in that: The newly added learning model in the step (S4) continuously updates the new model, and classifies and incrementally matches the quantified features of the wind turbine in a category manner.